DALLAS – Clinicians frequently perform bronchoalveolar lavage in ventilated trauma patients without radiologic evidence of pneumonia, according to a retrospective analysis.
Among 1,343 chest x-ray reports from 344 patients who all underwent bronchoalveolar lavage (BAL), there was no mention of infiltrates in 11% and no suspicion of pneumonia in 64%, according to a review that used natural language processing to sift through the reports.
Dr. Heather L. Evans
"Our indication for BAL includes chest x-ray infiltrates or a change in chest x-rays, so I was very surprised to see that there were so many BALs done without an infiltrate mentioned in the chest x-ray report," said lead author Dr. Heather L. Evans, a trauma and acute care surgeon and surgical intensivist at the University of Washington in Seattle. "I think that this may be something of a soft call when providers are concerned that the patient has increasing secretions, increasing oxygenation, and worsening sepsis of unknown etiology. Perhaps the chest x-ray is not as firm and fast a rule as we are led to believe."
Indeed, the Centers for Disease Control and Prevention removed the chest x-ray from its new surveillance definition for what is now termed adult ventilator-associated events. The new definition, expected to be implemented in 2013, is not intended for clinical management, leaving physicians in a quandary when making a clinical diagnosis of VAP. Enter natural language processing, a tool that is increasingly being applied in radiology as part of machine learning to aid in-text analysis of radiology reports (Med. Image Anal. 2012 [doi:10.1016/j.media.2012.02.005]).
The investigators used natural language processing coding methods to code 1,343 chest x-ray reports from the day prior, day of, and day after BAL among 344 trauma patients ventilated for more than 48 hours at a level 1 trauma center. Two specially trained reviewers coded the reports using the chest x-ray element from the Clinical Pulmonary Infection Score (CPIS) as "no infiltrate," "diffuse infiltrate or atelectasis," or "focal infiltrate" and scored the reports on a three-point scale for suspicion of pneumonia as "no suspicion," "suspicion," or "probable pneumonia."
The CPIS classifier had a 90% overall accuracy, 93% specificity, 86% sensitivity, and 85% positive predictive value. The suspicion classifier achieved comparable results of 85%, 89%, 78%, and 78%, respectively.
As expected, localized infiltrate was significantly more common in reports from BAL-positive than BAL-negative patients (13% vs. 9%), while no infiltrate was significantly more common in those from BAL-negative patients (15.3% vs. 11.5%). However, 1,013 chest x-ray reports, or 75.4% of the data, fell in-between with diffuse infiltrate or atelectasis and had a 50-50 chance of being diagnosed as VAP, Dr. Evans said at the annual meeting of the Surgical Infection Society.
"Failure to discriminate diffuse infiltrate defines the group where culture data is most useful," she observed.
Radiology reports noting any suspicion of pneumonia were significantly more common in positive-BAL than in negative-BAL patients (45.6% vs. 28%), while reports with no suspicion of pneumonia were significantly more common in BAL-negative patients (68% vs. 60%).
Still, 430 (50%) of the 856 chest x-ray reports with no suspicion of VAP were in patients with BAL-positive results, Dr. Evans pointed out.
To sort out the implications of this finding, the investigators stratified the CPIS data by time and discovered that differences between the BAL-positive and -negative groups regarding the presence or absence of infiltrates were statistically significant only on chest x-ray reports from the day after BAL (P = .004).
"Considering the timing of this chest x-ray report information is absolutely crucial and something we will definitely incorporate in the future," she said, adding that future work will involve evaluation of coded chest x-ray report content in VAP risk assessment.
Dr. E. Patchen Dellinger
Invited discussant Dr. Addison K. May, chief of trauma and surgical critical care at Vanderbilt University in Nashville, Tenn., questioned whether the authors were surprised by the findings given that chest x-ray readings and BAL results correlate only about 40% of the time, and asked why the authors chose to include the radiology report from the day after BAL. Dr. Evans said the lack of correlation wasn’t surprising and that chest x-ray report language will be incorporated, along with other available clinical values, into their VAP risk assessment model.
"To exclude the chest x-ray information is to ignore a fundamental piece of diagnostic data that clinicians use all the time," she added. "As much as we don’t like to rely on the chest x-ray, I’m currently doing a qualitative study of the diagnosis of ventilator-associated pneumonia at my institution, and I can tell you in the 15 interviews I’ve done, every single person says the chest x-ray is a fundamental piece that they rely on to make the diagnosis.